Google introduced Gemini 4 Argon on September 30, calling it the company's next era of frontier intelligence. The model is built to hold deep reasoning across long, multi-step workflows and is rolling out first to a cohort of trusted cyber defenders through Google's Fairwind Program. Its headline spec is an industry-leading one million token output limit, up from 64K on the previous generation. Introductory API pricing is 2 dollars per million input tokens and 10 dollars per million output tokens, with cached input tokens discounted by 95 percent. After the introductory period, those rates rise to 4 dollars and 20 dollars respectively.
Argon posts frontier results on real-world software engineering and on enterprise knowledge work such as legal drafting and financial research. Partners report early gains. Box said the model connected evidence across a complete set of documents, including details buried in long filings, and LexisNexis saw strong performance on legal knowledge and analysis inside its Legal Intelligence Engine. Google also leans hard into defensive cybersecurity, where Argon can autonomously find, validate, and patch critical software vulnerabilities. Internal teams put it to work in Antigravity to speed up molecular modeling, migrate more than 700 training labs, and refactor a codebase of roughly six million lines.
For founders, the practical takeaway is scope. A one million token output ceiling makes long-horizon agentic builds, large refactors, and document-heavy research workflows realistic in a single pass, while the introductory rates bring frontier reasoning cheap enough to prototype with. Access remains the catch. Argon is not generally available yet. Google is expanding availability in phases, says it is engaged in the US government's voluntary pre-release model access process, and will keep gathering tester feedback on guardrails before opening the model up to developers, enterprises, and consumers.